奖励引导的辅助防空反导自主作战决策研究  

Research into Autonomous Combat Decision-making in Assist Air Defense and Anti-missile Based on Reward Guiding

作　　者：韩兴豪 曹志敏[1] 刘家祺 李旭辉 HAN Xing-hao;CAO Zhi-min;LIU Jia-qi;LI Xu-hui(Jiangsu Automation Research Institute,Lianyungang 222061,China)

机构地区：[1]江苏自动化研究所,江苏连云港222061

出　　处：《舰船电子对抗》2021年第3期26-30,40,共6页Shipboard Electronic Countermeasure

摘　　要：现代作战态势愈加复杂,人工智能的跨越发展为提高军事决策智能化水平提供了新思路。利用深度强化学习技术,发挥深度网络在态势特征提取方面的能力,结合强化算法对智能体决策方法的迭代与优化,实现了作战智能化。针对复杂的作战态势,提出了一种对战训练框架,为执行辅助防空反导任务的歼击机构建智能体,并利用奖励重塑的方法缓解稀疏奖励问题,探讨了全面实现军事决策智能化的道路。With the combat situation becoming more and more complicated,the spanning development of artificial intelligence provides a new thought to improve the intelligent level of military decision-making.This paper uses deep reinforcement learning technology,exerts the ability to extract the situation feature of deep neural network,combines reinforcement learning algorithm to iterate and optimize the strategy of intelligent agents,achieves combat intelligent,aiming at the complicated combat situation,introduces a combat training framework,constructs the intelligent agents for the fighter to perform the task assisting air defense and anti-missile;and uses reward remodeling method to alleviate the problem of sparse reward,discusses the way to realize the military decision-making intelligent fully.

关 键 词：现代作战 深度强化学习 军事决策 奖励重塑 防空反导 

分 类 号：E91[军事]